Fix DDP synchronization in training callbacks

Fix multiple callbacks that caused training to freeze/deadlock after each
epoch when using multi-GPU DDP training. The issue was that some processes
reached trainer.strategy.barrier() while others returned early, causing
infinite waits.

Callbacks fixed:
- UnifiedVizCallback: Changed early return pattern to use if block
- EpochEndEvaluationCallback: Added missing barrier, restructured to avoid
  early returns
- WandBVizCallback: Replaced early return with conditional block
- WandBVizCallbackWithPAFs: Same fix as WandBVizCallback

The correct pattern (used by working callbacks like CSVLoggerCallback,
MatplotlibSaver, TrainingControllerZMQ) is:

    if trainer.is_global_zero:
        # rank-0 only work (no early returns!)
    trainer.strategy.barrier()  # ALL ranks must reach this

Co-Authored-By: Claude Opus 4.5 <[email protected]>

Author: gitttt-1234